115 results about "Scene labeling" patented technology

The invention discloses an image scene labeling method based on deep learning. The method comprises the steps of establishing a scene image data set, constructing a convolutional neural network, training a model, and labeling an image. The scene image data set is used for training and testing a deep learning scene recognition model. According to the construction of the convolutional neural network, the model of the convolutional neural network for scene recognition is constructed. According to the training of the model, the scene recognition model is obtained by training the convolutional neural network. According to the labeling of the image, a scene labeling word of the image is obtained through the identification of the image in the model. The shortage of image scene labeling is solved,and the accuracy of image scene labeling is improved.

The invention discloses an image scene labeling method based on a conditional random field and a secondary dictionary study, comprising steps of performing superpixel area over-segmentation on a training set image, obtaining a superpixel over-segmentation area of each image, extracting the characteristics of each superpixel over-segmentation area, combining with a standard labeled image to construct a superpixel label pool, using the superpixel label tool to train a support vector machine classifier to calculate superpixel unary potential energy, calculating paired item potential energy of adjacent superpixels, in virtue of global classification statistic of the over-segmentation superpixel area in a training set, constructing a classifier applicable to a class statistic histogram as a classification cost, using the histogram statistic based on the sum of the sparse coders of the sparse representation of the key point characteristic in each class superpixel area as the high order potential energy of a CRF model, using two distinguishing dictionaries of a class dictionary and a shared dictionary to optimize the sparse coder through the secondary sparse representation, and updating the dictionary, the CRF parameters and the classifier parameters. The image scene labeling method improves the labeling accuracy.

The invention discloses a scene-based automatic driving simulation test evaluation service cloud platform and an application method thereof. The system invention is used for automatic driving simulation test and is provided with a client, a cloud and a background management terminal, the client transmits corresponding function operation requirements to the cloud through a function module interface, and the cloud drives corresponding function modules to execute corresponding functions according to the function operation requirements; and the background management terminal manages and maintainsthe cloud and the client. In platform application, the method comprises functional applications of data uploading, data fusion, scene extraction, scene labeling, scene analysis, scene generation, simulation analysis and test evaluation. The whole tool chain including the scene database, the scene extraction and labeling, the scene analysis, the scene generation, the simulation analysis, the test evaluation and the like is integrated to the cloud, unified scheduling, management and use of automatic driving simulation data and test evaluation resources are realized, and the test efficiency is improved.

The present invention provides a video scene labeling device and method, and relates to the film and television media field. According to the video scene labeling device and method, a computer can be utilized to sample the video fragments of a single scene to thereby obtain a plurality of single-frame images; then a convolutional neural network algorithm is utilized to extract an image feature vector of each single-frame image; and according to a recurrent neural network algorithm, the plurality of pre-stored video fragments with the video labels and the image feature vector of each single-frame image, the video fragments are labeled, so that the video scenes can be labeled automatically without needing the manual intervention, the time cost and the labor cost are saved, and the user operation experiences are high.

The embodiment of the invention provides a bullet screen information processing method and device, electronic equipment and a storage medium, relates to the technical field of artificial intelligence,and relates to a machine learning technology. The method comprises the steps: obtaining a to-be-processed image in image data; performing scene recognition on the to-be-processed image to determine ascene label to which the to-be-processed image belongs, wherein the scene label is used for representing a scene category to which the image belongs; and taking the selectable barrage information associated with the scene label as candidate barrage information, and displaying the candidate barrage information. According to the embodiment of the invention, the user is guided to improve the behavior of sending the barrage according to the identified scene label, and the utilization rate of the barrage information is improved; proper barrage information can be triggered in a targeted manner, thestep of using general barrage information of a platform is avoided, and the accuracy and timeliness of barrage information sending are improved.

The invention relates to a deep learning based raft cultivation remote sensing image scene labeling method, which comprises the following four steps: step one, a computer reads data; step two, a multi-scale pure convolutional neural network is built; step three, the network is trained; step four, image labeling is performed, and a final result map is acquired. The method overcomes deficiencies in the prior art, well solves a problem of raft cultivation remote sensing image scene labeling, is high in automation degree and labeling precision, and can greatly reduce the labor cost, so that the method can be applied to labeling of raft cultivation remote sensing images and has broad application prospects and values.

The invention discloses a video scene analysis method and a device based on a human body, and the method comprises the steps: collecting an image, carrying out the scene labeling of the image, extracting a human body feature vector of the image through a deep learning method, forming a training sample through the human body feature vector of the image and the labeled scene, and forming a trainingsample set; training a random forest model by utilizing the training set, and obtaining a scene analysis model after parameters of the random forest model are determined; and reading each frame of image of a video to be analyzed, extracting a human body feature vector of the frame of image by using a deep learning method, and calculating and outputting a scene classification result of each frame of image based on the input human body feature vector by using the scene analysis model. According to the method and the device, the scene type of the video can be accurately identified, and the requirements of video automatic analysis and editing on rapid and accurate scene type calculation are met.

The embodiment of the invention provides a video search method and device and a server. The video search method is applied to the server, and comprises the steps: obtaining a keyword inputted by a user; when the keywords contain preset condition keywords, searching videos containing first target scene labels matched with the keywords according to the attribute information of the videos, and takingthe videos as first target videos; obtaining first target timestamp information of a card segment corresponding to the first target scene label in the first target video; and determining the identification information of the first target video and the first target timestamp information as search results. As scene recognition is performed on the video in advance, and the recognition result is counted, the attribute information of each video is obtained, wherein the attribute information comprises the scene label of each scene of each video and the timestamp information of the card segment corresponding to each scene label in the video to which the scene label belongs; and the video of which the video content is most matched with the keyword can be found according to the matching degree ofthe keyword and the scene label.

The invention provides a voice recognition method and system, electronic equipment and a storage medium, and the method comprises the steps: obtaining training sample sets of different scenes, and enabling the training sample sets to comprise a plurality of training voices and text labels corresponding to the training voices; training a preset machine learning model according to the training sample sets of the different scenes to obtain semantic models corresponding to the different scenes; to-be-recognized voice is obtained, wherein the to-be-recognized voice carries a scene label; obtainingsemantic models corresponding to the scene labels from the semantic models corresponding to the different scenes; processing the to-be-recognized voice by utilizing the target semantic model to obtainan initial recognition result of the to-be-recognized voice; and performing calibration processing on the initial recognition result by using a preset language model to obtain a target recognition result of the to-be-recognized voice. According to the invention, the problems that targeted voice recognition cannot be carried out for a specific service scene of a user and the recognition accuracy is not high can be solved.

The invention provides an outdoor media advertisement intelligent delivery method and system, and the method comprises the steps: carrying out the scene labeling analysis of each outdoor media terminal according to the positioning information and scene information of each outdoor media terminal, and obtaining the label data of each outdoor media terminal; analyzing the advertisement delivery requirements of the advertiser, matching the advertisement delivery requirements of the advertiser with the label data of each outdoor media terminal, and screening out the outdoor media terminals meetingthe advertisement putting requirements of the advertiser; and delivering intelligent advertisements on the screened outdoor media terminal. The advertisement delivery requirements of the advertiser are matched with the label data of the outdoor media terminals, the outdoor media terminals meeting the delivery requirements are screened out, and therefore accurate and intelligent delivery of the outdoor media advertisements is achieved.

The invention discloses a model training method implemented based on a machine learning technology. The method comprises the steps of obtaining a to-be-trained content sample image and a to-be-trained style sample image; generating a to-be-trained simulation sample image according to the to-be-trained content sample image and the to-be-trained style sample image; obtaining a first prediction scene label and a first prediction style label of the to-be-trained simulation sample image through a to-be-trained image recognition model; obtaining a second prediction scene label and a second prediction style label of the to-be-trained style sample image through the to-be-trained image recognition model; and updating model parameters of the to-be-trained image recognition model according to the prediction label and a labeling label until model training conditions are met, and outputting the image recognition model. The invention further provides an image recognition method and device. According to the invention, more sample images belonging to a target domain are expanded by using the labeled image samples, the collection requirements of different scene data in the target domain are met, and the generalization ability of the image recognition model is improved.

The invention provides a scene recognition method and device, electronic equipment and a storage medium, and belongs to the technical field of imaging. The method comprises the steps that according toa preset rule, a current preview image is divided into N areas, and N is a positive integer larger than 1; utilizing a preset scene recognition model to perform scene recognition on the N areas respectively so as to determine a scene label corresponding to each area; determining a scene label corresponding to the current preview image according to the scene label corresponding to each area; and determining a target shooting mode according to the scene label corresponding to the current preview image. Therefore, through the scene recognition method, mutual interference of various image contents in the preview image is reduced, the scene recognition accuracy is improved, and the user experience is improved.

The invention relates to an image labeling method and device, a readable medium and electronic equipment, and relates to the technical field of image processing, the method comprises the following steps: inputting a to-be-labeled target image into a pre-trained image multi-classification model, obtaining a matching degree between a target image output by the image multi-classification model and each scene label in the plurality of scene labels, and obtaining a first preset number of feature maps extracted by the image multi-classification model, determining a second preset number of target scene tags in the plurality of scene tags according to the matching degree of each scene tag, and labeling the target image according to the first preset number of feature maps and the image binary classification model corresponding to each target scene tag. According to the invention, the target scene label is screened out by using the image multi-classification model, and the target image is labeled according to the image binary classification model corresponding to the target scene label, so that a plurality of scene labels can be labeled for the image, and the image labeling accuracy and calculation efficiency are improved.

The invention provides a test scene determination method and device, electronic equipment and a readable storage medium, and the method comprises the steps: determining at least one standard scene cluster according to a pre-calculated mapping relation between a standard parameter set corresponding to each standard scene and a scene label; calculating the distance between each obtained candidate scene and each standard scene cluster; and determining the candidate scene as a target test scene, wherein the distance between the candidate scene and each standard scene cluster is greater than a preset distance threshold value corresponding to the standard scene. According to the method and the device, the test scene of the information not in the standard scene cluster is determined from the candidate scenes, so that the test scene of the same type as the standard scene can be obtained not only by taking a value near the standard scene, the scene types of advantages and disadvantages of the automatic driving algorithm are enriched and determined, the simulation test can be comprehensively performed on the automatic driving algorithm, and the comprehensiveness and accuracy of subsequent testing of the to-be-tested automatic driving algorithm are ensured.

The invention relates to a voice emotion recognition method and device, a medium and electronic equipment, and belongs to the technical field of emotion recognition. The method comprises the steps: extracting various types of audio features of user voice when the user voice is received; matching the audio features with feature samples in an emotional feature library to obtain emotional labels corresponding to the feature samples matched with the audio features; constructing a feature tag matrix of the user voice based on the audio features and the emotion tags corresponding to the matched feature samples; inputting the feature tag matrix into a multi-emotion recognition model to obtain a plurality of emotion sets and scene tags corresponding to the emotion sets; and acquiring a scene tag matched with the voice scene of the user voice so as to determine an emotion set corresponding to the matched scene tag as the recognized user voice emotion. According to the invention, various potential emotions can be efficiently and accurately recognized from the voice.

The invention discloses a vehicle collision detection method based on machine learning. The vehicle collision detection method based on machine learning comprises the following steps: acquiring a velocity vector, an acceleration vector, an angular velocity vector and longitude and latitude; preprocessing the acquired speed data respectively; calculating new vector data through the preprocessed data and acquiring a road scene label; composing the calculated data into an input vector; calculating collision probabilities under different models through input vectors and calculating comprehensive collision probabilities; acquiring the category of the input vector through a preset method and judging whether the comprehensive collision probability value and the category of the input vector are abnormal or not; and marking the input vector as a collision vector or a non-collision vector. By integrating supervised learning and unsupervised learning algorithms, the cost of machine learning is reduced; collision detection is carried out by using a plurality of trained classifiers, and collision detection under a plurality of collision scenes is covered by adopting a mode of deep mining of a plurality of dimensions and a plurality of collision scenes, so that the accuracy rate and the recall rate of collision detection are improved.

The invention provides a video editing method, system and device based on scene recognition and a storage medium. The method comprises the steps of: extracting each frame of an original video as a first picture, generating a first picture set, arranging the first pictures in the first picture set according to the sequence of the pictures in the original video, and forming a frame chain table; cutting the pictures in the first picture set to remove markers to obtain second pictures, and generating a second picture set; sequentially adding a lens label to each second picture in the second picture set according to a lens recognition model, and adding a scene label; editing the second picture set according to a preset target frame number, the target scene label and a preset scene priority sequence, and sequentially outputting third pictures to obtain a third picture set; and synthesizing all the third pictures in the third picture set according to the sequence in the frame chain table, andoutputting the edited video. According to the invention, videos can be automatically clipped in batches, the work of artificial video synthesis is replaced, the operation cost is greatly saved, and the operation efficiency is effectively improved.

The invention discloses a real scene three-dimensional semantic reconstruction method and device based on deep learning and a storage medium, relates to the technical field of remote sensing surveying and mapping geographic information, and solves the problem of inaccurate multi-scene labeling in the prior art. The method comprises: obtaining anaerial image; carrying out semantic segmentation on the aerial image, and determining a pixel probability distribution diagram; performing motion structure recovery on the aerial image, and determining a camera pose of the aerial image; performing depth estimation on the aerial image, and determining a depth map of the aerial image; and performing semantic fusion on the pixel probability distribution map, the camera pose and the depth map to determine a three-dimensional semantic model. Thus, high-precision segmentation is realized under the conditions of more scene objects, serious stacking and the like is realized; and in a large-scale scene, the performance of the depth estimation network is not affected, stable and accurate estimation can be carried out in various scenes, and compared with other traditional three-dimensional reconstruction algorithms, the semantic three-dimensional reconstruction algorithm constructed by the invention has the advantage that the calculation speed is increased.

The invention provides an image shooting method and device, a pan-tilt camera and a storage medium and relates to the technical field of computer vision. The method comprises steps of obtaining a second image collected by a second camera in a camera preview or photographing process; performing scene identification on the second image to obtain a scene label of the second image; and determining shooting parameters of the first camera according to the scene label of the second image, and controlling the first camera to perform image shooting based on the shooting parameters. According to the method, scene identification is carried out on the second image collected by the second camera, the scene label of the second image can be obtained, the resolution of the second camera is small, operation efficiency of scene identification can be improved, the shooting parameters are determined based on the scene label, the first camera can shoot high-definition image data according to the shooting parameters, and the shooting experience of a user is improved.

The invention relates to the artificial intelligence technology, and discloses a rescue scene image identification method. The method comprises the following steps: using a trailer scene image set anda power-on scene image set to train an initial image identification model to obtain a first target detection model and a second target detection model; aggregating the first target detection model and the second target detection model into a parallel detection model, and preliminarily judging whether the scene label to which the to-be-identified rescue scene image belongs is a trailer scene or apower-on scene or not by utilizing the parallel detection model, and inputting the to-be-identified rescue scene image into the first target detection model or the second target detection model according to the scene label to which the to-be-identified rescue scene image belongs for image identification. The invention further provides a rescue scene image identification device and a computer readable storage medium. In addition, the invention also relates to a blockchain technology, and the trailer scene image set and the power-on scene image set can be stored in the blockchain node. Accordingto the invention, the efficiency of identifying the rescue scene image can be improved.

The embodiment of the invention discloses a household commodity recommendation method, a client and a server. The method comprises the following steps: providing an image sample library associated with home furnishing, wherein the image sample library comprises a scene image sample and a commodity image sample associated with the scene image sample, wherein the household commodity represented by the commodity image sample is located in a scene represented by the scene image sample, the scene image sample is provided with a scene label. The method comprises the following steps: receiving an indoor shot image sent by a client, and identifying the scene label of the shot image from the shot image based on a pre-trained neural network; determining a target scene image sample matched with the shot image from an image sample library according to the identified scene label of the shot image; and feeding back the household commodity represented by the commodity image sample associated with thetarget scene image sample to the client. According to the technical scheme provided by the invention, the home shopping experience of the user can be improved while the home shopping cost is reduced.

The invention discloses a method and device for determining a function page started by an electronic book application and a storage medium. The method comprises the steps: collecting user historical operation behavior data related to function page operation and user historical scene label data corresponding to user historical operation behaviors after the electronic book application is started; obtaining user stay function page preference data according to the user historical operation behavior data, and establishing a first corresponding relation between the user stay function page preferencedata and the user historical scene label data; after it is monitored that the e-book application is started, acquiring user real-time scene label data, and inquiring user stay function page preference data matched with the user real-time scene label data according to the first corresponding relation; and determining a function page started by the e-book application according to the user stay function page preference data. According to the method, the function page started by the e-book application can be determined in a targeted manner according to different scenes, and the effect of startingthe function page by thousands of people is achieved.

The invention discloses a model training and scene recognition method and device, equipment and a medium. The method comprises the following steps: training a first scene label and standard cross entropy loss of a sample image during the training of a scene recognition model, obtaining the parameters of a core feature extraction layer and a global information feature extraction layer through the training, then, according to a feature map output by the LCS module of each level and a loss value obtained by pixel-by-pixel calculation of the first scene label of the sample image, training a weight parameter of the LCS module of each level, and finally, training to obtain a parameter of a full-connection decision layer of the scene recognition model. Therefore, the scene recognition model has the high-richness feature extraction capability, scene recognition is performed based on the scene recognition model, and the accuracy of scene recognition is greatly improved.

The embodiment of the invention provides an advertisement putting method, an advertisement putting server, a client and an advertisement putting system. The method comprises the following steps that an advertisement putting server receives a label pulling request; the advertisement putting server returns a scene label corresponding to the video identifier of the target video to the client; the advertisement putting server receives the advertisement request, wherein the advertisement request is sent by the client in advance before the target appearing moment; the advertisement putting server returns an advertisement protocol matched with the scene label identifier in the advertisement request to the client; and the client is used for calculating coordinate points for exploring advertisement rendering according to the position and the size of a target object issued by the advertisement protocol, and displaying advertisement materials in each target putting frame based on the calculated coordinate points, the rendering mode and the putting time information. The relevancy between the advertisement and the video content is increased, the advertisement putting position is relatively flexible and diversified, and the effect that the advertisement moves along the path of the target object can be achieved.

The invention discloses a short video label determination method, system and device and a storage medium, and the method comprises the steps: obtaining the audio information of a first short video, and carrying out the video-audio analysis of the first short video, and obtaining a first audio label; acquiring key frame information of a first short video, and performing video content analysis on the first short video to obtain a first scene label, a first object label and a first character label; acquiring title information, video description information and subtitle information of the first short video, and performing video semantic analysis on the first short video to obtain a first semantic tag; and performing weight decision analysis according to the first audio tag, the first scene tag, the first object tag, the first character tag and the first semantic tag to generate a first short video tag. The short video tag generation efficiency is improved, and the accuracy, comprehensiveness and reliability of the short video tag are also improved. The method can be widely applied to the technical field of video processing.

The invention discloses a city defense and control image detection method based on asynchronous federated learning. The method comprises the steps that a cloud server initializes a global model; each end device carries out scene labeling on city defense and control image data, divides the city defense and control image data into a training set and a test set, and screens out a data set which is close to the sample distribution of the test set from the training set as a to-be-trained set; a global model is obtained from the cloud server, and the to-be-trained set is locally trained to obtain a local model; each end device uploads the local model to the cloud server after homomorphic encryption; and the cloud server performs global model training on the local model by adopting an asynchronous federation calculation strategy to obtain an updated global model.

The invention relates to the field of automatic driving, and particularly discloses an automatic driving regulation and control algorithm optimization method, which comprises the following steps: constructing M test scenes according to a scene label tree; performing simulation operation on an automatic driving regulation and control algorithm based on the M test scenes to obtain M simulation results; for each test scene in the M test scenes, evaluating simulation results corresponding to the test scenes according to an evaluation algorithm corresponding to the test scenes to obtain M evaluation results; obtaining a mapping scene tag tree according to the M evaluation results and the M test scenes; and determining problem characteristics of the automatic driving regulation and control algorithm according to the mapping scene tag tree, and optimizing the automatic driving regulation and control algorithm according to the problem characteristics. According to the embodiment of the invention, the analysis efficiency of a large number of simulation results is improved, meanwhile, the problem characteristics of the automatic driving regulation and control algorithm are given, and the automatic driving regulation and control algorithm is optimized based on the problem characteristics.

The invention provides a display scene recognition method and device, a model training method and device, equipment, a storage medium and a computer program product, relates to the technical field of artificial intelligence, in particular to the technical field of computer vision and deep learning, and can be applied to scenes such as image processing and image recognition. According to the specific implementation scheme, the method comprises the steps of obtaining feature vectors of a to-be-recognized image and obtaining a base library feature vector set; determining at least two candidate feature vectors from the bottom library feature vector set based on the feature vector of the to-be-recognized image and the similarity coefficient of each feature vector in the bottom library feature vector set; performing threshold judgment on the similarity coefficients of the at least two candidate feature vectors to obtain a target feature vector; and determining the display scene of the to-be-recognized image based on the display scene label corresponding to the target feature vector. The display scene is identified according to a mode of performing threshold judgment on the similarity coefficient of the candidate feature vector, so that the accuracy is ensured, the labeling cost is reduced, and the identification efficiency is improved.

The invention provides a method for automatically matching a communication contact person. A contact person database of a communication party is established according to history communication information of the communication party; a session scene label value currently corresponding to a session scene label is obtained through collection according to the preset session scene label; and a communication person corresponding to the session scene label value is obtained from the contact person database through matching. The technical problem that an existing mode of obtaining the contact person by manually inputting contact person information for searching and matching is tedious in operation and low in efficiency is solved. The communication contact person is rapidly and automatically obtained through matching; the communication contact person obtained from the contact person database through matching according to the session scene label value is fully combined with the current session scene, so the method has relatively high pertinence and individuation and is relatively high in intelligence.